Bionic ripple shape design method of composite material functional structure

Abstract

The invention provides a bionic corrugation form design method of a composite material functional structure, and belongs to the technical field of composites.The bionic corrugation form design method comprises the steps that three-dimensional scanning and dynamic capturing are conducted on various fish body surface corrugation structures through a high-speed underwater camera system, and a fish body surface corrugation database is established; a bionic ripple basic parameter set is divided into a nanoscale level, a micron level and a millimeter level according to the scale to construct a multi-scale ripple characteristic matrix, a game model is constructed to realize double-layer collaborative optimization, and a hydrodynamic performance test of a bionic composite material test piece is performed in an underwater fluid experiment device through an orthogonal experiment design method. Directional jitter increasing and expanding processing is carried out on experimental data to generate a rich training data set, an iterative control threshold value is calculated to judge design precision, closed-loop feedback optimization of theoretical prediction and experimental verification is achieved, and the technical problem that accurate bionics of the fish body surface multi-scale ripple structure are difficult to achieve is solved.

Description

Technical Field

[0001] This invention belongs to the field of composite material technology, and more specifically, relates to a biomimetic corrugated morphology design method for functional structures of composite materials. Background Technology

[0002] In the field of composite material surface structure design for underwater devices, biomimetic technology improves hydrodynamic performance by simulating the microstructure of biological surfaces. Traditional biomimetic design methods are mainly based on macroscopic observation and simplified modeling of the surface morphology of marine organisms such as shark skin and dolphin skin. They simulate biological surface features by fabricating regular groove structures or uniform microtextures. These methods are widely used in biomimetic drag-reducing coatings, biomimetic surface treatment technologies, and bio-inspired material design. However, the biomimetic process is often limited to single-scale morphological replication, lacking a comprehensive understanding and systematic reproduction of the multi-level structural features of biological surfaces. Existing biomimetic ripple design technologies typically employ static geometric parameter extraction and fixed fabrication processes, failing to accurately capture and reproduce the dynamic changes and complex spatial distribution characteristics of fish surface ripples at different scale levels, particularly the organic correlation and synergistic mechanism between nanoscale surface textures, micrometer-scale groove structures, and millimeter-scale macroscopic ripples. In practical biomimetic design, due to the lack of precise quantitative analysis of the morphological parameters of fish body surface ripples and a unified modeling method for multi-scale features, traditional biomimetic technologies struggle to achieve high-fidelity replication of the complex surface structures of biological prototypes. This results in a significant gap between the actual performance of biomimetic materials and the superior properties of the biological prototypes. In other words, existing technologies face the technical challenge of accurately mimicking the multi-scale ripple structure of fish body surfaces. Summary of the Invention

[0003] In view of this, the present invention provides a biomimetic wavy morphology design method for functional structures of composite materials, which can solve the technical problem in the prior art that it is difficult to achieve accurate biomimetic multi-scale wavy structures on the surface of fish.

[0004] This invention is implemented as follows: This invention provides a biomimetic ripple morphology design method for composite material functional structures, comprising: extracting surface ripple morphology parameters based on a fish surface ripple database, including ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle, to establish a biomimetic ripple basic parameter set; dividing the biomimetic ripple basic parameter set into nanoscale surface textures, micrometer-scale groove structures, and millimeter-scale macroscopic ripples according to ripple scale, constructing a multi-scale ripple feature matrix; using an adaptive morphology prediction model to evaluate the hydrodynamic performance of the multi-scale ripple feature matrix, calculating the drag coefficient and lift coefficient; and based on the multi-view graph coloring problem... The topology optimization algorithm is used for optimization design, dividing the surface of the composite material into multiple adjacent regions and assigning the optimal combination of corrugation parameters to each region; an upper-level game model and a lower-level game model are constructed, and the corrugation morphology design parameters are determined through bi-level game optimization; composite material specimens are prepared, and the fluid dynamics performance is tested using orthogonal experimental design method to obtain experimental data; the experimental data are subjected to directional jitter amplification processing, the degree of deviation and data distribution density are calculated, and the iterative control threshold is obtained; it is determined whether the iterative control threshold meets the design accuracy requirements. If it does, the final parameters are output; otherwise, the results are fed back to the performance evaluation step for re-evaluation.

[0005] The step of establishing a database of fish surface ripples involves using a high-speed underwater camera system to perform three-dimensional scanning of the surface ripple structure of various fish species. The system consists of a circular array of multiple high-resolution underwater cameras, equipped with underwater lighting equipment and a water flow control device. By precisely controlling the shooting angle and lighting conditions, it achieves all-round three-dimensional imaging of the fish's surface in swimming state. The camera system uses a high-speed shooting capability of 10,000 frames per second to capture the dynamic changes of the surface ripples of fish in different swimming postures.

[0006] During the data collection process, three common nearshore fish species were selected as research subjects, including yellow croaker, ribbonfish, and pomfret. By controlling environmental parameters in the tank, such as water temperature, salinity, and flow rate, the natural living environment of the fish was simulated to maintain their normal swimming behavior and body surface morphology. Based on this, key parameters such as the amplitude, wavelength, phase difference, and tilt angle of the ripples were recorded.

[0007] Among them, the nanoscale surface texture refers to the microscopic surface structure with a scale range of 1 to 1000 nm, the micrometer-level trench structure refers to the mesoscopic surface morphology with a scale range of 1 to 1000 μm, the millimeter-level macroscopic ripples refer to the macroscopic geometric features with a scale range of 1 to 100 mm, and the multi-scale ripple feature matrix is ​​used to characterize the combination relationship of ripple morphology parameters at different scale levels.

[0008] The adaptive morphology prediction model is specifically based on a deep neural network with a Transformer-XL architecture, containing 12 encoder layers and 8 decoder layers. Each layer contains 512 hidden units and 16 multi-head attention mechanisms. The memory length parameter is dynamically adjusted according to the temporal feature length of the multi-scale ripple feature matrix and the surface area of ​​the composite material.

[0009] Before training the adaptive morphology prediction model, the training dataset is established by collecting surface ripple parameters of 1,000 different fish species as input samples, obtaining the corresponding drag coefficient and lift coefficient as label data through computational fluid dynamics simulation, and dividing the dataset into training, validation and test sets in an 8:1:1 ratio.

[0010] Specifically, the training steps of the adaptive morphological prediction model involve using the Adam optimizer for gradient descent training, setting the learning rate to 0.0001, the batch size to 32, and the number of training epochs to 500. An early stopping strategy is used to prevent overfitting, and the model performance is evaluated on the validation set every 10 epochs. Training is stopped when the validation loss does not decrease for 20 consecutive epochs.

[0011] Specifically, the multi-objective topology optimization algorithm based on the graph coloring problem involves meshing the surface of the composite material into several hexagonal regions, with each region serving as a node in the graph. Adjacent regions are connected by edges, and the combination of ripple parameters serves as the color of the nodes. Constraints require that the colors of adjacent nodes satisfy fluid continuity requirements. The objective function is to minimize the overall fluid resistance, and a genetic algorithm is used to solve the graph coloring optimization problem.

[0012] The objective function of the upper-level game model includes the reciprocal of the drag coefficient, the logarithm of the Reynolds number, the negative square root of the ratio of ripple amplitude to ripple wavelength, and the sine function of ripple phase difference. The objective function of the lower-level game model includes the reciprocal of the maximum stress, the cube root of the product of the composite material's elastic modulus and the material thickness, the negative exponential of the ratio of ripple amplitude to material density, and the cosine function of the ripple tilt angle.

[0013] The underwater fluid experimental device includes an underwater fluid circulation system, a composite material specimen clamping device, a multi-dimensional force sensor, a flow velocity control system, and a data acquisition system. The underwater fluid circulation system provides a stable water flow environment with a flow velocity control range of 0.1 to 10.0 m / s and a flow velocity control accuracy of 0.01 m / s. The multi-dimensional force sensor measures the resistance and lift force experienced by the specimen with a measurement accuracy of 0.001 N.

[0014] The orthogonal experimental design method specifically employs a four-factor, three-level orthogonal array for experimental design. The four factors are ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle. Each factor has three levels: high, medium, and low, resulting in a total of nine experimental combinations. During the experiment, the prepared composite material specimen is first mounted on a clamping device. The angle between the specimen surface and the water flow direction is adjusted. The underwater fluid circulation system is started, and the target flow velocity is set. Data acquisition begins after the flow field stabilizes. Each experimental group lasts for 300 seconds, with a sampling frequency of 1000 Hz.

[0015] Specifically, the directional jitter augmentation process employs a Gaussian noise-based data augmentation method to add Gaussian noise with a mean of 0 and a standard deviation of 10% of the experimental values ​​to the experimental drag coefficient and experimental lift coefficient, thereby generating an augmented experimental dataset that is 10 times larger than the original data.

[0016] The deviation is calculated using the root mean square error method, which calculates the square root of the average of the squared differences between each data point in the augmented experimental dataset and the corresponding theoretical prediction value. The data distribution density is calculated using the kernel density estimation method, which calculates the probability density distribution of the augmented experimental dataset in different value intervals.

[0017] Furthermore, prior to the fluid dynamics performance evaluation step, there is also a step of establishing an adaptive adjustment function for the ripple parameters. The adaptive adjustment function for the ripple parameters is used to adjust the memory length parameter of the adaptive morphology prediction model. Based on three data points—the temporal correlation strength of the multi-scale ripple feature matrix, the geometric complexity of the composite material surface, and the Reynolds number of the flow field—a memory adjustment factor value is obtained.

[0018] Specifically, when the memory adjustment factor value is in the range of 0 to 0.3, a linearly increasing weight adjustment function is used to adjust the memory length parameter to 1.2 times the base value; when the memory adjustment factor value is in the range of 0.3 to 0.7, a constant weight adjustment function is used to maintain the memory length parameter at the base value; and when the memory adjustment factor value is in the range of 0.7 to 1.0, an exponentially decaying weight adjustment function is used to adjust the memory length parameter to 0.8 times the base value.

[0019] This invention establishes a database of fish surface ripples and a multi-scale ripple feature matrix based on a high-speed underwater camera system, achieving precise quantification and parameterization of fish surface ripple structures from the nanometer to the millimeter scale. This solves the problems of inaccurate extraction of biological prototype features and low biomimetic fidelity in traditional biomimetic technologies. The invention employs 3D reconstruction algorithms and stereoscopic vision technology to capture the dynamic changes of fish surface ripples in swimming states. It combines an adaptive morphology prediction model with deep learning to model the complex relationships between multi-scale ripple parameters. A two-layer game optimization algorithm is used to achieve collaborative optimization design of biomimetic ripple parameters. A complete technical chain is established, from precise observation of biological prototypes, accurate extraction of feature parameters, multi-scale correlation modeling to biomimetic structural optimization design. This enables biomimetic composite materials to reproduce the multi-level structural features and excellent hydrodynamic performance of fish surface ripples with high fidelity. In summary, this invention solves the technical problem of accurately mimicking the multi-scale ripple structure of fish surfaces. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a schematic diagram of a high-speed underwater camera system.

[0022] Figure 3 This is a schematic diagram of the adaptive morphology prediction model structure.

[0023] Figure 4 This is a schematic diagram of the underwater fluid experimental setup. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0025] like Figure 1 The diagram shows a flowchart of a biomimetic corrugated morphology design method for a composite material functional structure provided by this invention. This method includes the following steps:

[0026] S01. Based on the established fish body surface ripple database, extract the body surface ripple morphology parameters of various fish species. The body surface ripple morphology parameters include ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle, and establish a set of biomimetic ripple basic parameters.

[0027] S02. Divide the biomimetic ripple basic parameter set into three levels according to the ripple scale: nanoscale surface texture, micrometer-level groove structure and millimeter-level macro ripple, and construct a multi-scale ripple feature matrix.

[0028] S03. An adaptive morphological prediction model is used to evaluate the fluid dynamics performance of the ripple amplitude, ripple wavelength, ripple phase difference and ripple tilt angle in the multi-scale ripple feature matrix, and the drag coefficient and lift coefficient are calculated.

[0029] S04. A multi-objective topology optimization algorithm based on the graph coloring problem optimizes the ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle, divides the composite material surface into multiple adjacent regions, and assigns the optimal combination of ripple parameters to each region.

[0030] S05. Construct an upper-level game model with the goal of minimizing fluid resistance and a lower-level game model with the goal of maximizing material strength. Optimize the two-level game to determine the final corrugation design parameters.

[0031] S06. Prepare composite material specimens according to the corrugation morphology design parameters, and conduct fluid dynamic performance tests in an underwater fluid experimental device using orthogonal experimental design methods to obtain experimental drag coefficient and experimental lift coefficient.

[0032] S07. Perform directional jitter augmentation processing on the experimental drag coefficient and experimental lift coefficient to generate an augmented experimental dataset. Calculate the degree of deviation between the augmented experimental dataset and the theoretical prediction value and the data distribution density to obtain the iterative control threshold.

[0033] S08. Determine whether the iterative control threshold meets the design accuracy requirements. When the iterative control threshold is greater than the preset threshold, feed the expanded experimental dataset back to step S03 to re-evaluate the fluid dynamics performance. When the iterative control threshold is less than or equal to the preset threshold, output the final ripple shape design parameters.

[0034] The specific implementation method for establishing a fish surface ripple database involves using a high-speed underwater camera system to perform three-dimensional scanning of the surface ripple structure of various fish species. This system consists of a circular array of multiple high-resolution underwater cameras, equipped with underwater lighting and a flow control device. By precisely controlling the shooting angle and lighting conditions, it achieves omnidirectional stereoscopic imaging of the fish's surface during swimming. The camera system employs a high-speed shooting capability of 10,000 frames per second to capture the dynamic changes in the surface ripples of fish under different swimming postures, obtaining complete ripple morphology data from head to tail, ensuring data continuity and integrity. During data acquisition, three common and easily obtainable safe nearshore fish species were selected as research subjects: yellow croaker, ribbonfish, and pomfret. These fish are widely distributed in nearshore waters and are easy to catch and raise. They exhibit typical surface ripple characteristics and good experimental adaptability. By controlling tank environmental parameters such as water temperature, salinity, and flow velocity, the natural living environment of the fish was simulated to maintain their normal swimming behavior and surface morphology. Based on this, key parameters such as ripple amplitude, wavelength, phase difference, and tilt angle were recorded. The 3D reconstruction algorithm uses stereo vision technology to convert a sequence of 2D images taken from multiple angles into an accurate 3D ripple model. Through steps such as feature point matching, depth estimation, and surface reconstruction, it generates high-precision 3D point cloud data of the fish body surface. Then, it uses surface fitting and ripple analysis algorithms to extract quantitative values ​​of ripple parameters from the point cloud data, establishes a complete parameterized ripple feature description, and finally forms a comprehensive fish body surface ripple database containing original image data, 3D reconstruction models, and ripple parameter values, providing reliable biological basis data support for subsequent biomimetic ripple design.

[0035] Among them, nanoscale surface texture refers to microscopic surface structures with a scale range of 1–1000 nm, micrometer-scale groove structures refer to mesoscopic surface morphologies with a scale range of 1–1000 μm, and millimeter-scale macroscopic ripples refer to macroscopic geometric features with a scale range of 1–100 mm. A multi-scale ripple feature matrix is ​​used to characterize the combination relationship of ripple morphology parameters at different scale levels. Each element in the matrix corresponds to a ripple parameter value at its corresponding scale level. The ripple amplitude is derived from the measured peak height values ​​in the fish body surface ripple database and is used for topology optimization calculations in step S04. The ripple wavelength is derived from the measured peak spacing values ​​in the fish body surface ripple database and is used for calculating the objective function of the upper-level game model in step S05. The ripple phase difference is derived from the measured phase relationship values ​​of adjacent ripples in the fish body surface ripple database and is used for setting the constraint conditions of the lower-level game model in step S05. The ripple tilt angle is derived from the measured ripple direction angle values ​​in the fish body surface ripple database and is used for the preparation of composite material specimens in step S06.

[0036] The adaptive morphology prediction model is structured as a deep neural network based on the Transformer-XL architecture, comprising a 12-layer encoder and an 8-layer decoder. Each layer contains 512 hidden units and 16 multi-head attention mechanisms. The memory length parameter is dynamically adjusted based on the temporal feature length of the multi-scale ripple feature matrix and the surface area of ​​the composite material. The steps for establishing the training dataset for the adaptive morphology prediction model specifically include collecting surface ripple parameters from 1000 different fish species as input samples, obtaining the corresponding drag coefficient and lift coefficient as label data through computational fluid dynamics simulation, dividing the dataset into training, validation, and test sets in an 8:1:1 ratio, normalizing the ripple amplitude, wavelength, phase difference, and tilt angle, and constructing a temporal input format. The training steps of the adaptive morphological prediction model specifically include gradient descent training using the Adam optimizer, with a learning rate of 0.0001, a batch size of 32, 500 training epochs, an early stopping strategy to prevent overfitting, evaluation of model performance on the validation set every 10 epochs, and stopping training when the validation loss does not decrease for 20 consecutive epochs. Finally, the prediction accuracy of the model on the test set reaches over 95%.

[0037] The multi-objective topology optimization algorithm based on the graph coloring problem meshes the surface of the composite material into several hexagonal regions, each region serving as a node in the graph. Adjacent regions are connected by edges, and the combination of ripple parameters serves as the color of the nodes. The constraint condition requires that the colors of adjacent nodes satisfy the fluid continuity requirement. The objective function is to minimize the overall fluid resistance. The graph coloring optimization problem is solved using a genetic algorithm.

[0038] The objective function of the upper-level game model includes the reciprocal of the drag coefficient, the logarithm of the Reynolds number, the negative square root of the ratio of ripple amplitude to ripple wavelength, and a sine function of the ripple phase difference. The objective function of the lower-level game model includes the reciprocal of the maximum stress, the cube root of the product of the composite material's elastic modulus and material thickness, the negative exponential of the ratio of ripple amplitude to material density, and a cosine function of the ripple tilt angle. The coupling term is the ratio function of ripple amplitude to ripple wavelength. The constraints of the upper-level game model include ripple amplitude range constraints, ripple wavelength range constraints, and ripple phase difference range constraints. The constraints of the lower-level game model include material strength constraints, material thickness constraints, and ripple tilt angle constraints. The objective function of the upper-level game model is used to optimize fluid dynamics performance. The inputs include the drag coefficient, Reynolds number, ripple amplitude, ripple wavelength, and ripple phase difference, and the output is a comprehensive evaluation index of fluid performance. The objective function of the lower-level game model is used to optimize the structural strength of the material. The inputs include the maximum stress, the elastic modulus of the composite material, the material thickness, the material density, and the corrugation tilt angle. The output is a comprehensive evaluation index of structural strength.

[0039] The orthogonal experimental design method employs a four-factor, three-level orthogonal array for experimental design. The four factors are ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle. Each factor has three levels: high, medium, and low, resulting in a total of nine experimental combinations. The underwater fluid experimental apparatus includes an underwater fluid circulation system, a composite material specimen clamping device, a multi-dimensional force sensor, a flow velocity control system, and a data acquisition system. The underwater fluid circulation system provides a stable water flow environment with a flow velocity control range of 0.1–10.0 m / s and a flow velocity control accuracy of 0.01 m / s. The composite material specimen clamping device fixes the specimen and maintains the relative position of the specimen surface with respect to the water flow direction. The multi-dimensional force sensor measures the resistance and lift forces experienced by the specimen with a measurement accuracy of 0.001 N. The data acquisition system records the force sensor output signals in real time and calculates the experimental resistance coefficient and experimental lift coefficient. During the experiment, the prepared composite material specimen was first installed on the clamping device, the angle between the specimen surface and the water flow direction was adjusted, the underwater fluid circulation system was started and the target flow rate was set, and data acquisition began after the flow field stabilized. Each set of experiments lasted for 300 seconds and the sampling frequency was 1000Hz. After the experiment was completed, the specimen surface was cleaned and the next set of experimental parameters was changed.

[0040] The directional jitter augmentation process employs a Gaussian noise-based data augmentation method, adding Gaussian noise with a mean of 0 and a standard deviation of 10% of the experimental values ​​to the experimental drag and lift coefficients, generating an augmented experimental dataset ten times the size of the original data. The deviation is calculated using the root mean square error method, taking the square root of the average of the squared differences between each data point in the augmented experimental dataset and its corresponding theoretical prediction. The data distribution density is calculated using kernel density estimation, determining the probability density distribution of the augmented experimental dataset across different value intervals. The iterative control threshold is calculated by weighting the deviation and data distribution density, with the weighting coefficients determined based on the experimental accuracy requirements.

[0041] An adaptive adjustment function for the ripple parameter is established. This function is used to adjust the memory length parameter of the adaptive morphological prediction model. Based on three data points—the temporal correlation strength of the multi-scale ripple feature matrix, the geometric complexity of the composite material surface, and the Reynolds number of the flow field—a memory adjustment factor value is obtained. When the memory adjustment factor value is in the range of 0 to 0.3, a linearly increasing weight adjustment function is used to adjust the memory length parameter to 1.2 times the base value. When the memory adjustment factor value is in the range of 0.3 to 0.7, a constant weight adjustment function is used to maintain the memory length parameter at the base value. When the memory adjustment factor value is in the range of 0.7 to 1.0, an exponentially decaying weight adjustment function is used to adjust the memory length parameter to 0.8 times the base value.

[0042] The underwater fluid experimental setup achieves its technical advantages by precisely controlling experimental conditions and employing a high-precision measurement system to obtain reliable hydrodynamic performance data of composite materials, providing accurate experimental verification for iterative optimization of ripple morphology design. The setup utilizes a closed-loop fluid circulation system to avoid interference from environmental factors in open water experiments. The high-precision measurement capabilities of the multi-dimensional force sensors ensure the accuracy of drag and lift coefficient measurements, while the high sampling frequency of the data acquisition system guarantees the integrity and continuity of experimental data. The application of orthogonal experimental design methods allows for the acquisition of maximum information with minimal experiments, significantly improving experimental efficiency and data utilization. The directional jitter augmentation processing method, through a reasonable noise addition strategy, maintains the authenticity of the experimental data while increasing its diversity, providing richer training samples for subsequent iterative optimization and effectively improving the convergence speed and final accuracy of the design method.

[0043] The specific implementation methods of the above steps are described in detail below. The specific implementation method of step S01 is to first construct a high-speed underwater camera system, such as... Figure 2 As shown, the system consists of 8-12 high-resolution underwater cameras arranged in a circular array, equipped with LED underwater lighting and a controllable water flow device. By precisely controlling the shooting angle and lighting conditions, it achieves omnidirectional stereoscopic imaging of the fish's body surface during swimming. The camera system uses a high-speed shooting capability of 10,000 frames per second to capture the dynamic changes in the ripples on the fish's body surface under different swimming postures, obtaining complete ripple morphology data from head to tail. During data acquisition, three common nearshore fish species—yellow croaker, ribbonfish, and pomfret—were selected as research subjects. Environmental parameters in the tank, such as water temperature (18-25℃), salinity (3.0-3.5%), and flow velocity (0.2-2.0 m / s), were controlled to simulate the fish's natural living environment, ensuring they maintained normal swimming behavior and body surface morphology. The 3D reconstruction algorithm uses stereo vision technology to convert a sequence of 2D images captured from multiple angles into a precise 3D ripple model. Through steps such as feature point matching, depth estimation, and surface reconstruction, it generates high-precision 3D point cloud data of the fish's body surface, achieving a point cloud accuracy of 0.1 mm. Then, surface fitting and ripple analysis algorithms are used to extract quantitative values ​​of ripple parameters from the point cloud data, establishing a complete parametric ripple feature description. The purpose of this step is to provide reliable biological foundational data support for subsequent biomimetic ripple design, obtaining accurate morphological parameters of the fish's body surface ripples through high-precision 3D scanning technology.

[0044] The specific implementation of step S02 involves dividing the biomimetic corrugation basic parameter set obtained in step S01 into hierarchical levels according to the corrugation scale. Nanoscale surface textures refer to microscopic surface structures with a scale range of 1–1000 nm; micrometer-scale trench structures refer to mesoscopic surface morphologies with a scale range of 1–1000 μm; and millimeter-scale macroscopic corrugations refer to macroscopic geometric features with a scale range of 1–100 mm. A multi-scale corrugation feature matrix is ​​established to characterize the combination relationship of corrugation morphology parameters at different scale levels. Each element in the matrix corresponds to the value of a corrugation parameter at the corresponding scale level. The ripple amplitude is derived from crest height measurements in the fish surface ripple database, ranging from 0.01 to 50 mm. The ripple wavelength is derived from crest spacing measurements in the same database, ranging from 0.1 to 200 mm. The ripple phase difference is derived from phase relationship measurements between adjacent ripples in the database, ranging from 0 to 360°. The ripple tilt angle is derived from ripple direction angle measurements in the database, ranging from 0 to 90°. This step employs multi-scale analysis theory to systematically classify complex biological surface structures according to spatial scales. The aim is to establish a structured ripple feature description system, providing a standardized input data format for subsequent optimization calculations.

[0045] The specific implementation of step S03 involves using an adaptive morphological prediction model based on the Transformer-XL architecture to evaluate the hydrodynamic performance of the ripple parameters in the multi-scale ripple feature matrix. For example... Figure 3 As shown, Figure 3 As shown, the prediction model comprises a 12-layer encoder and an 8-layer decoder, with each layer containing 512 hidden units and 16 multi-head attention mechanisms. The memory length parameter is dynamically adjusted based on the temporal feature length of the multi-scale ripple feature matrix and the surface area of ​​the composite material. The base memory length is set to 128, with an adjustment range of 100–160. The model input consists of normalized ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle parameters, and the output consists of the corresponding predicted drag coefficient and lift coefficient. This step utilizes the attention mechanism and long short-term memory capabilities of deep learning to capture the complex nonlinear relationships and temporal dependencies between ripple parameters, aiming to quickly and accurately predict the fluid dynamics performance under different ripple configurations, thus avoiding extensive computational fluid dynamics simulations.

[0046] The specific implementation of step S04 is to optimize the corrugation parameters using a multi-objective topology optimization algorithm based on the graph coloring problem. The composite material surface is meshed into several hexagonal regions with side lengths ranging from 5 to 20 mm. Each region serves as a node in the graph, with connecting edges between adjacent regions. The corrugation parameter combination is used as the node color. Constraints require that the colors of adjacent nodes satisfy fluid continuity requirements, i.e., the difference in corrugation amplitude between adjacent regions does not exceed 20%, the difference in corrugation wavelength does not exceed 15%, the corrugation phase difference changes continuously, and the difference in corrugation tilt angle does not exceed 10°. The objective function is to minimize the overall fluid resistance. A genetic algorithm is used to solve the graph coloring optimization problem, with a population size of 100, a crossover probability of 0.8, a mutation probability of 0.1, and 500 generations. This step transforms the continuous corrugation parameter optimization problem into a discrete graph coloring problem, using combinatorial optimization theory to ensure parameter continuity between adjacent regions. The aim is to achieve the optimal configuration of local corrugation parameters while maintaining surface geometric continuity.

[0047] The specific implementation of step S05 involves constructing a two-layer game optimization model to determine the final corrugation design parameters. The upper-layer game model aims to minimize fluid resistance. Its objective function includes the reciprocal of the resistance coefficient, the logarithm of the Reynolds number, the negative square root of the ratio of corrugation amplitude to corrugation wavelength, and a sine function of the corrugation phase difference. Constraints include a corrugation amplitude range of 0.01–50 mm, a corrugation wavelength range of 0.1–200 mm, and a corrugation phase difference range of 0–360°. The lower-layer game model aims to maximize material strength. Its objective function includes the reciprocal of the maximum stress, the cube root of the product of the composite material's elastic modulus and material thickness, the negative exponential of the ratio of corrugation amplitude to material density, and a cosine function of the corrugation tilt angle. Constraints include a material strength of not less than 200 MPa, a material thickness of 1–10 mm, and a corrugation tilt angle of 0–90°. The coupling term is a function of the ratio of corrugation amplitude to corrugation wavelength, with coupling coefficients set to 0.1–0.5. This step uses Nash equilibrium theory from game theory to handle the multi-objective conflict between fluid performance and structural strength. It achieves Pareto optimal solution through iterative solution, with the aim of finding the best balance between fluid dynamic performance and material structural strength.

[0048] The specific implementation of step S06 involves preparing composite material specimens based on the corrugation morphology design parameters determined in step S05, using a four-factor, three-level orthogonal array L9(3). 4 The experimental design involved four factors: ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle. Each factor was set at three levels: high, medium, and low, resulting in a total of nine experimental combinations. Figure 4As shown, the underwater fluid experimental setup includes an underwater fluid circulation system, a composite material specimen clamping device, a multi-dimensional force sensor, a flow velocity control system, and a data acquisition system. The flow velocity control range is 0.1–10.0 m / s, with a flow velocity control accuracy of 0.01 m / s. The multi-dimensional force sensor has a measurement accuracy of 0.001 N, and the data acquisition system has a sampling frequency of 1000 Hz. During the experiment, the prepared composite material specimen is mounted on the clamping device, the angle between the specimen surface and the water flow direction is adjusted, the underwater fluid circulation system is started, and the target flow velocity is set. Data acquisition begins after the flow field stabilizes, and each experimental group lasts for 300 seconds. This step employs orthogonal experimental design theory to obtain the maximum amount of information with the fewest experiments, aiming to verify the accuracy of theoretical predictions through a standardized experimental procedure.

[0049] The specific implementation of step S07 involves directional jitter augmentation of the experimental drag coefficient and lift coefficient. A Gaussian noise-based data augmentation method is used to add Gaussian noise with a mean of 0 and a standard deviation of 10% of the experimental value to the experimental data, generating an augmented experimental dataset ten times the size of the original data. The deviation is calculated using the root mean square error method, taking the square root of the average of the squared differences between each data point in the augmented experimental dataset and its corresponding theoretical prediction value, with a deviation threshold set to 5%. Data distribution density is calculated using kernel density estimation with a Gaussian kernel function and a bandwidth parameter set to 0.1, calculating the probability density distribution of the augmented experimental dataset across different value intervals. The iterative control threshold is calculated by a weighted average of the deviation and data distribution density, with weighting coefficients determined based on experimental accuracy requirements, typically set to 0.6 and 0.4. This step employs the Monte Carlo method in statistics to increase the diversity of the data samples, aiming to improve the robustness of model training and provide richer validation data for iterative optimization.

[0050] The specific implementation of step S08 involves determining whether the iterative control threshold meets the design accuracy requirements. When the iterative control threshold is greater than the preset threshold of 0.05, the expanded experimental dataset is fed back to step S03 for a re-evaluation of fluid dynamics performance, updating the weight parameters of the adaptive morphology prediction model, and simultaneously establishing an adaptive adjustment function for the corrugation parameters to adjust the model's memory length parameter. This adjustment function calculates the memory adjustment factor value based on three data: the temporal correlation strength of the multi-scale corrugation feature matrix, the geometric complexity of the composite material surface, and the Reynolds number of the flow field. When the memory adjustment factor value is in the range of 0 to 0.3, a linearly increasing weight adjustment function is used to adjust the memory length parameter to 1.2 times the base value. When the memory adjustment factor value is in the range of 0.3 to 0.7, a constant weight adjustment function is used to maintain the memory length parameter at the base value. When the memory adjustment factor value is in the range of 0.7 to 1.0, an exponentially decaying weight adjustment function is used to adjust the memory length parameter to 0.8 times the base value. When the iterative control threshold is less than or equal to the preset threshold of 0.05, the final corrugation morphology design parameters are output. This step employs feedback control theory to achieve an adaptive optimization process, aiming to continuously improve the accuracy of design parameters through iterative corrections until they meet engineering application requirements.

[0051] The detailed structure of the adaptive morphological prediction model is based on the Transformer-XL architecture, designed for long sequence modeling problems and effectively handling the temporal dependencies of multi-scale ripple features. The main body of the model comprises an encoder-decoder structure. The encoder consists of 12 identical layers, each containing a multi-head self-attention sub-layer and a feedforward neural network sub-layer. Each sub-layer employs residual connections and layer normalization. The multi-head self-attention mechanism includes 16 attention heads, each with a dimension of 32, for a total dimension of 512. The decoder consists of 8 layers, including the two sub-layers of the encoder and an encoder-decoder attention sub-layer. Positional encoding uses a relative positional encoding method, which can better handle input sequences of different lengths. The memory mechanism is a core innovation of Transformer-XL. By maintaining a memory state in each layer to store the hidden state of the previous segment, the model can utilize longer contextual information. The memory length parameter is dynamically adjusted according to the complexity of the input data, with a base length of 128 and an adjustment range of 100–160.

[0052] The detailed steps for establishing the training dataset for the adaptive morphological prediction model are as follows: First, surface ripple parameters of 1000 different fish species were collected as input samples. These data originated from the fish surface ripple database established in step S01, covering different ecological types of fish, including freshwater fish, marine fish, benthic fish, and mid-to-upper-level fish. Computational fluid dynamics simulation software was used to numerically simulate each set of ripple parameters, obtaining the corresponding drag coefficient and lift coefficient as label data. The simulation employed a Reynolds-averaged Navier-Stokes equation solver, and the k-ω shear stress transport model was selected as the turbulence model. The mesh size was controlled between 1 million and 5 million, and the convergence accuracy was set to 10. -6 The 1000 sets of data were randomly divided into a training set of 800 sets, a validation set of 100 sets, and a test set of 100 sets in an 8:1:1 ratio to ensure that the ripple data of different types of fish were evenly distributed across the subsets. The ripple amplitude, wavelength, phase difference, and tilt angle were normalized to their maximum and minimum values, scaling the numerical range to between 0 and 1. A time-series input format was constructed, with each sample containing a sequence of ripple parameters for 32 time steps.

[0053] The model was trained using the Adam optimizer with gradient descent, a learning rate of 0.0001, a batch size of 32, and 500 training epochs. A learning rate decay strategy was employed, multiplying the learning rate by 0.9 every 100 epochs. Early stopping was used to prevent overfitting, and model performance was evaluated on the validation set every 10 epochs. Training stopped when the validation loss did not decrease for 20 consecutive epochs. The loss function was a weighted combination of mean squared error loss and smoothed L1 loss, with weights of 0.7 and 0.3. Gradient clipping was used during training to prevent gradient explosion, with a clipping threshold of 1.0. The final trained model achieved a prediction accuracy of over 95% on the test set, with drag coefficient prediction errors controlled within 3% and lift coefficient prediction errors controlled within 5%.

[0054] The key technical ideas of this invention include multi-scale biomimetic ripple parameter extraction technology, graph coloring-based topology optimization algorithm, two-layer game optimization model, and adaptive morphology prediction technology. The multi-scale biomimetic ripple parameter extraction technology, through a high-speed underwater camera system and 3D reconstruction algorithm, can accurately acquire multi-level ripple structure parameters from nanometer to millimeter scales on the surface of fish. Compared with traditional single-scale measurement methods, this technology can comprehensively reflect the multi-scale characteristics of biological surfaces, providing a more complete and accurate biological data foundation for biomimetic design, and significantly improving the biorealism and functional effectiveness of biomimetic ripple designs. The graph coloring-based topology optimization algorithm transforms the continuous parameter optimization problem into a discrete combinatorial optimization problem. It uses coloring theory in graph theory to ensure the continuity constraint of ripple parameters between adjacent regions. Compared with traditional continuous optimization methods, this algorithm can better handle complex boundary conditions and geometric constraints, avoid the local optima problem that easily occurs in traditional methods, and ensure the engineering feasibility of the optimization results. The two-level game-theoretic optimization model transforms the multi-objective optimization problem into an equilibrium problem within a game theory framework by establishing an upper-level game for fluid performance optimization and a lower-level game for structural strength optimization. Compared to traditional weighted multi-objective optimization methods, this model can better balance the conflicting relationships between different objectives, avoid the subjectivity and arbitrariness of weight coefficient settings, and obtain Pareto optimal solutions with stronger theoretical support and engineering guidance. The adaptive morphology prediction technology, based on the Transformer-XL deep learning architecture, captures the long-term dependencies and complex nonlinear mappings between ripple parameters through memory and attention mechanisms. Compared to traditional empirical formulas and simplified models, this technology can significantly improve the accuracy and efficiency of fluid dynamics performance prediction and reduce the reliance on extensive computational fluid dynamics simulations.

[0055] The synergistic effect of these key technological approaches forms a complete biomimetic corrugated design optimization system. Multi-scale parameter extraction provides high-quality input data for the optimization algorithm, graph coloring topology optimization ensures the geometric continuity of the design results, a two-layer game model achieves multi-objective balance, and adaptive prediction technology provides fast and accurate performance evaluation. These four technologies support and work together. Compared with existing biomimetic design methods, this collaborative technology system can simultaneously guarantee the bio-realism, geometric feasibility, optimal performance, and computational efficiency of the design. It fundamentally solves key technical problems in traditional methods such as inaccurate biological feature extraction, difficulty in handling optimization constraints, strong subjectivity in multi-objective trade-offs, and low efficiency in performance evaluation, providing a systematic technical solution for the biomimetic design of composite functional structures.

[0056] It should be noted that this invention also solves the following three key technical problems:

[0057] The first technical challenge is how to accurately acquire and effectively utilize the biomimetic characteristic parameters of fish surface ripples. Traditional biomimetic designs often rely on static observation or simple measurements, making it difficult to capture the true dynamic characteristics of fish surface ripples during swimming, resulting in inaccurate biomimetic parameters. This invention establishes a high-speed underwater camera system, employing a high-speed shooting capability of 10,000 frames per second and multiple high-resolution cameras arranged in a circular array to achieve omnidirectional stereoscopic imaging of the fish surface during swimming, accurately capturing the dynamic changes of ripples. Simultaneously, it utilizes stereoscopic vision technology and 3D reconstruction algorithms to convert the multi-angle captured 2D image sequence into an accurate 3D ripple model. Through steps such as feature point matching, depth estimation, and surface reconstruction, high-precision 3D point cloud data is generated, thereby establishing a complete biomimetic ripple database containing key parameters such as ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle.

[0058] The second technical challenge is achieving coordinated optimization design of multi-scale corrugation features. The corrugated structure of composite material surfaces involves multiple scales, including nanometer, micrometer, and millimeter levels. Traditional optimization methods struggle to handle the coupling relationships and constraints between different scales. This invention employs a multi-objective topology optimization algorithm based on a graph coloring problem. The composite material surface is meshed into several hexagonal regions, each region serving as a node in the graph. The combination of corrugation parameters is used as the node's color. By constraining the colors of adjacent nodes to meet fluid continuity requirements, coordinated configuration of corrugation parameters across different regions is achieved. Combined with an adaptive morphology prediction model based on the Transformer-XL architecture, utilizing its deep neural network structure of a 12-layer encoder and an 8-layer decoder, the hydrodynamic performance of different corrugation parameter combinations is accurately predicted, providing a reliable performance evaluation basis for the optimization algorithm.

[0059] The third technical challenge is establishing a reliable experimental verification and iterative optimization mechanism. There is often a discrepancy between the theoretical predictions and actual performance of biomimetic wave pattern designs, lacking an effective experimental verification and feedback optimization mechanism. This invention constructs a complete underwater fluid experimental device, including a high-precision multi-dimensional force sensor and a stable fluid circulation system, obtaining reliable experimental data through orthogonal experimental design methods. An innovative directional jitter augmentation processing method is proposed, generating an augmented experimental dataset by adding appropriate Gaussian noise, maintaining data authenticity while increasing diversity. An iterative control threshold judgment mechanism based on the degree of deviation and data distribution density is established. When the experimental results deviate significantly from the theoretical predictions, the augmented experimental dataset is fed back to the performance evaluation step to retrain the model, forming a closed-loop optimization process to ensure the accuracy and reliability of the final design parameters.

[0060] Specifically, the principle of this invention is as follows: The key to achieving precise biomimicry of multi-scale ripple structures on the surface of fish lies in establishing a high-fidelity replication system covering the entire process from biological prototype to biomimetic product. This system achieves dynamic and precise capture of the ripple structure on the surface of fish through high-speed underwater camera technology and 3D reconstruction algorithms, overcoming the limitations of traditional static observation methods that cannot acquire dynamic features of biological surfaces. A circular array composed of multiple high-resolution underwater cameras, combined with a high-speed shooting capability of 10,000 frames per second, can capture the instantaneous changes in the surface ripples of fish under different swimming postures. High-precision 3D point cloud data is generated through stereo vision technology and feature point matching algorithms, achieving accurate quantification of key parameters such as ripple amplitude, wavelength, phase difference, and tilt angle. The multi-scale ripple feature matrix uniformly represents the ripple parameters at different scale levels, establishing a mathematical correlation model between nanoscale surface textures, micrometer-level groove structures, and millimeter-level macroscopic ripples, enabling biomimetic design to simultaneously consider structural features and interaction relationships at multiple scale levels. An adaptive morphology prediction model based on the Transformer-XL architecture captures the complex nonlinear relationships between surface ripple parameters of fish through a deep learning network. The adaptive adjustment mechanism for the memory length parameter dynamically optimizes the model structure according to the complexity of ripple characteristics in different fish species, ensuring accurate modeling and prediction of surface ripple features for various fish. A two-layer game-theoretic optimization model, through hierarchical processing of upper-layer fluid performance optimization and lower-layer structural strength optimization, ensures that the biomimetic ripple parameters meet fluid dynamics performance requirements while maintaining material structural reliability. A graph-shading topology optimization algorithm ensures the continuous distribution and local optimization of ripple parameters on the composite material surface, avoiding discontinuities in biomimetic structures. An iterative verification and data augmentation mechanism, through closed-loop feedback between experimental testing and theoretical prediction, continuously corrects and improves the biomimetic design parameters, ensuring a high degree of consistency between the final biomimetic product and the biological prototype's performance characteristics.

[0061] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0062] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0063] The specific implementation of step S02 is to construct a multi-scale ripple feature matrix, which is specifically represented as follows:

[0064]

[0065] In the formula, M is the multi-scale ripple feature matrix; A ijLet be the j-th ripple parameter at the i-th scale level, where i = 1, 2, 3 correspond to nanometer, micrometer, and millimeter scales respectively, and j = 1, 2, 3, 4 correspond to ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle respectively. Matrix element normalization is performed using the following formula:

[0066]

[0067] In the formula, These are the normalized matrix elements; It is the minimum value of the j-th parameter at the i-th scale level; It represents the maximum value of the j-th parameter at the i-th scale level.

[0068] The specific implementation of step S03 is to use an adaptive morphological prediction model to calculate the drag coefficient and lift coefficient. The drag coefficient prediction formula is as follows:

[0069]

[0070] The formula for predicting the lift coefficient is:

[0071]

[0072] In the formula, C D C is the drag coefficient; L f is the lift coefficient; drag f is the drag coefficient prediction function; lift This is the lift coefficient prediction function, which is implemented by a deep neural network based on the Transformer-XL architecture.

[0073] The specific implementation of step S04 is the same as described above, and will not be repeated in detail here.

[0074] The specific implementation of step S05 is to construct a two-layer game optimization model, where the objective function of the upper-layer game model is:

[0075]

[0076] In the formula, F upper The objective function of the upper-level game model is: w1, w2, w3, and w4 are the weight coefficients of the upper-level model, with values ​​ranging from 0.3 to 0.5, 0.1 to 0.2, 0.2 to 0.4, and 0.1 to 0.3, respectively; Re is the Reynolds number. The objective function of the lower-level game model is:

[0077]

[0078] In the formula, F lowerσ represents the objective function value of the lower-level game model; v1, v2, v3, and v4 are the weight coefficients of the lower-level model, with values ​​ranging from 0.4 to 0.6, 0.2 to 0.3, 0.1 to 0.2, and 0.1 to 0.2, respectively; max ρ is the maximum stress; E is the elastic modulus of the composite material; t is the material thickness; ρ is the material density. The expression for the coupling term is:

[0079]

[0080] In the formula, C couple γ is the numerical value of the coupling term; γ is the coupling coefficient, with a value ranging from 0.1 to 0.5.

[0081] The parameter is obtained as follows: Re is obtained through calculation, and the calculation formula is: Where ρ f σ is the fluid density, v is the fluid velocity, L is the characteristic length, and μ is the dynamic viscosity; max The data were obtained using finite element analysis; E represents the inherent properties of the composite material, obtained through material mechanics tests; t represents the design parameters, ranging from 1 to 10 mm; and ρ represents the density of the composite material, obtained through density testing.

[0082] The specific implementation method of step S06 is the same as described above, and will not be repeated in detail here.

[0083] The specific implementation of step S07 involves performing directional jitter amplification processing on the experimental data. The Gaussian noise addition formula is as follows:

[0084]

[0085] In the formula, This represents the drag coefficient after expansion; The increased lift coefficient; The drag coefficient is the one measured experimentally. The lift coefficient was measured experimentally. and The mean is 0 and the variance is 1. and Gaussian distributed noise, where The deviation is calculated using the root mean square error formula:

[0086]

[0087] In the formula, RMSE is the root mean square error; n is the number of samples in the augmented dataset; This is the i-th augmented experimental data; Let be the i-th theoretical predicted value. The data distribution density is calculated using kernel density estimation:

[0088]

[0089] In the formula, is the estimated probability density function; h is the bandwidth parameter, with a value of 0.1; K is the Gaussian kernel function. x represents the location of the point to be estimated; x i Let be the i-th data point; u is the standardized variable. The formula for calculating the iterative control threshold is:

[0090]

[0091] In the formula, T control The threshold is used for iterative control; α and β are weighting coefficients, typically set to 0.6 and 0.4, respectively; x mean This represents the data mean.

[0092] The specific implementation of step S08 is to establish an adaptive adjustment function for the ripple parameters, and the formula for calculating the memory adjustment factor is:

[0093]

[0094] In the formula, R factor Memory regulator; TSC represents the strength of time-series correlation; TSC max GCC represents the maximum temporal correlation strength; GCC is the geometric complexity. max Re represents the maximum geometric complexity. max This represents the maximum Reynolds number. The memory length adjustment function is:

[0095] When 0≤R factor When <0.3,

[0096] When 0.3≤R factor When L < 0.7, memory =L base ;

[0097] When 0.7≤R factor When ≤1.0,

[0098] Where, L memory L is the adjusted memory length. base The base memory length is set to 128.

[0099] The parameter acquisition method is as follows: TSC is calculated using the autocorrelation function, and the calculation formula is: Where N is the sequence length, k is the number of lag steps, and its value ranges from 1 to 10, y i For the i-th time series data point; GCC is obtained by fractal dimension calculation, and the surface geometric complexity is calculated by box counting method, with the following formula: Where N(ε) is the number of boxes with side length ε, and ε is the size of the box.

[0100] The multi-scale ripple feature matrix is ​​established based on hierarchical representation theory. By systematically classifying complex biological surface structures according to spatial scale, a standardized description of ripple features at different levels is achieved. Compared to traditional single-parameter description methods, this matrix can comprehensively reflect the multi-scale hierarchical structure of biological surfaces, providing a structured data input format for subsequent optimization algorithms and significantly improving the accuracy and convergence speed of ripple parameter optimization. Normalization ensures that parameters at different scales have the same numerical range, avoiding optimization bias caused by differences in parameter magnitude.

[0101] The design principle of the objective function of the two-level game model is based on the Nash equilibrium theory in game theory. In the upper objective function, the reciprocal term of the drag coefficient aims to minimize fluid drag, the logarithmic term of the Reynolds number considers the influence of flow field characteristics, the negative square root term of the ratio of ripple amplitude to wavelength reflects the geometric scale effect, and the sine function term of the ripple phase difference reflects the periodic influence of the phase relationship on fluid performance. In the lower objective function, the reciprocal term of the maximum stress aims to maximize structural strength, the cube root term of the product of elastic modulus and thickness considers the material stiffness characteristics, the negative exponential term of the ratio of ripple amplitude to density reflects the lightweight design requirements, and the cosine function term of the ripple tilt angle reflects the influence of geometric angle on structural performance. Compared with the traditional weighted multi-objective optimization method, this two-level game model can better balance the conflict between fluid performance and structural strength, avoid the subjectivity of weight coefficient setting, and the obtained Pareto optimal solution has stronger theoretical support and engineering guidance significance.

[0102] The principle of Gaussian noise addition in directional jitter augmentation is based on the Monte Carlo method in statistics. By adding random noise with certain statistical characteristics, it simulates the uncertainty and measurement error of experimental data, thereby expanding the scale and diversity of the training dataset. Compared with traditional repeated experimental data collection methods, this method can significantly reduce experimental costs and time while maintaining data authenticity. The root mean square error calculation provides a quantitative accuracy evaluation index, the kernel density estimation method can accurately describe the probability distribution characteristics of the data, and the weighted calculation of the iterative control threshold comprehensively considers the dual constraints of prediction accuracy and data distribution, providing a reliable convergence criterion for model iterative optimization.

[0103] The design principle of the adaptive adjustment function for ripple parameters is based on adaptive control theory. The memory adjustment factor, by comprehensively considering three key factors—temporal correlation strength, geometric complexity, and flow field Reynolds number—can dynamically reflect the complexity of the input data. The piecewise memory length adjustment strategy employs different adjustment strategies under different complexity conditions: linear increment is suitable for low complexity to enhance learning ability, constant maintenance is suitable for a stable state under medium complexity, and exponential decay is suitable for high complexity to prevent overfitting. Compared with the traditional method of fixed memory length, this adaptive adjustment mechanism can dynamically optimize model parameters according to data characteristics, significantly improving the adaptability and prediction accuracy of the prediction model under different operating conditions.

[0104] These mathematical models and formulas constitute the core theoretical foundation of precise biomimetic design. Through the multi-scale ripple feature matrix M, a complete digital representation of biological surface structures is achieved, enabling biomimetic design to accurately capture features at all levels from nanometer to millimeter, and normalizing the formulas. This ensures a unified quantification standard for parameters at different scales. The upper-level objective function in the two-level game model.

[0105] and lower-level objective function

[0106] Mathematical rigor ensures an optimal balance between fluid performance and structural strength, where the reciprocal of the drag coefficient is used. Directly optimize drag reduction effect, ripple geometry ratio term It reflects the golden ratio characteristics of biological surfaces, and the reciprocal of the maximum stress term. Structural safety was ensured. Directional jitter augmentation was performed using a Gaussian noise model. The kernel density estimation function simulates random perturbations in a real-world environment. Accurately describes the probability distribution characteristics of experimental data, root mean square error

[0107] A quantitative accuracy evaluation standard is provided. The memory adjustment factor in the adaptive adjustment function. Through time series correlation and geometric complexity The quantitative calculations enable intelligent adjustment of model parameters. This precise biomimetic design method, through rigorous constraints and optimization of the mathematical model, can accurately reproduce the performance of efficient fluids in nature. Compared with traditional empirical biomimetic methods, the composite material surface designed by this method can maximize drag reduction while maintaining structural strength, providing scientific and technical support for the efficient and energy-saving operation of underwater devices.

[0108] It should be noted that the variables involved in this invention are explained in detail in Table 1.

[0109] Table 1. Variable Explanation Table

[0110]

[0111]

[0112] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A biomimetic corrugated morphology design method for functional structures of composite materials, used in underwater device applications, characterized in that... include: Based on the fish body surface ripple database, surface ripple morphology parameters were extracted, including ripple amplitude, ripple wavelength, ripple phase difference, and ripple tilt angle, to establish a set of basic parameters for biomimetic ripples. The set of basic parameters for biomimetic ripples was divided into nanoscale surface texture, micrometer-scale groove structure, and millimeter-scale macroscopic ripples according to the ripple scale, and a multi-scale ripple feature matrix was constructed. An adaptive morphology prediction model is used to evaluate the hydrodynamic performance of the multi-scale ripple feature matrix and calculate the drag coefficient and lift coefficient. The optimization design is based on a multi-objective topology optimization algorithm for graph coloring problem. The surface of the composite material is divided into multiple adjacent regions, and the optimal combination of corrugation parameters is assigned to each region. An upper-level game model and a lower-level game model are constructed, and the corrugation morphology design parameters are determined through two-level game optimization. Composite material specimens were prepared, and their fluid dynamic performance was tested using orthogonal experimental design methods to obtain experimental data. The experimental data is subjected to directional jitter amplification processing to calculate the degree of deviation and data distribution density, and the iterative control threshold is obtained. It is then determined whether the iterative control threshold meets the design accuracy requirements. If it does, the final parameters are output; otherwise, the data is fed back to the performance evaluation step for re-evaluation.

2. The biomimetic corrugated morphology design method for composite material functional structures according to claim 1, characterized in that, The steps to establish a database of fish surface ripples are as follows: a high-speed underwater camera system is used to perform three-dimensional scanning of the surface ripple structure of various fish species. The system consists of a circular array of multiple high-resolution underwater cameras, equipped with underwater lighting equipment and a water flow control device. By precisely controlling the shooting angle and lighting conditions, the system achieves all-round three-dimensional imaging of the fish surface in swimming states. The camera system uses a high-speed shooting capability of 10,000 frames per second to capture the dynamic changes of the surface ripples of fish in different swimming postures.

3. The biomimetic corrugated morphology design method for composite material functional structures according to claim 2, characterized in that, During the data collection process, three common nearshore fish species were selected as research subjects, including yellow croaker, ribbonfish, and pomfret. By controlling environmental parameters in the tank, such as water temperature, salinity, and flow rate, the natural living environment of the fish was simulated to maintain their normal swimming behavior and body surface morphology. Based on this, key parameters such as the amplitude, wavelength, phase difference, and tilt angle of the ripples were recorded.

4. The biomimetic corrugated morphology design method for composite material functional structures according to claim 3, characterized in that, The nanoscale surface texture refers to the microscopic surface structure with a scale range of 1 to 1000 nm, the micrometer-level trench structure refers to the mesoscopic surface morphology with a scale range of 1 to 1000 μm, the millimeter-level macroscopic ripples refer to the macroscopic geometric features with a scale range of 1 to 100 mm, and the multi-scale ripple feature matrix is ​​used to characterize the combination relationship of ripple morphology parameters at different scale levels.

5. The biomimetic corrugated morphology design method for composite material functional structures according to claim 4, characterized in that, The adaptive morphology prediction model is specifically based on a deep neural network with a Transformer-XL architecture, containing 12 encoder layers and 8 decoder layers. Each layer contains 512 hidden units and 16 multi-head attention mechanisms. The memory length parameter is dynamically adjusted according to the temporal feature length of the multi-scale ripple feature matrix and the surface area of ​​the composite material.

6. The biomimetic corrugated morphology design method for composite material functional structures according to claim 5, characterized in that, Before training the adaptive morphology prediction model, the training dataset is established by collecting surface ripple parameters of 1,000 different fish species as input samples and obtaining the corresponding drag coefficient and lift coefficient as label data through computational fluid dynamics simulation. The dataset is then divided into training, validation, and test sets in an 8:1:1 ratio.

7. The biomimetic corrugated morphology design method for composite material functional structures according to claim 6, characterized in that, The training steps of the adaptive morphological prediction model are as follows: gradient descent training is performed using the Adam optimizer, with a learning rate of 0.0001, a batch size of 32, and 500 training epochs. An early stopping strategy is used to prevent overfitting. The model performance is evaluated on the validation set every 10 epochs, and training is stopped when the validation loss does not decrease for 20 consecutive epochs.

8. The biomimetic corrugated morphology design method for composite material functional structures according to claim 7, characterized in that, The multi-objective topology optimization algorithm based on the graph coloring problem specifically involves meshing the surface of the composite material into several hexagonal regions, with each region serving as a node in the graph. Adjacent regions are connected by edges, and the combination of ripple parameters serves as the color of the nodes. Constraints require that the colors of adjacent nodes satisfy the fluid continuity requirement. The objective function is to minimize the overall fluid resistance, and a genetic algorithm is used to solve the graph coloring optimization problem.

9. The biomimetic corrugated morphology design method for composite material functional structures according to claim 8, characterized in that, The objective function of the upper-level game model includes the reciprocal of the drag coefficient, the logarithm of the Reynolds number, the negative square root of the ratio of ripple amplitude to ripple wavelength, and the sine function of ripple phase difference. The objective function of the lower-level game model includes the reciprocal of the maximum stress, the cube root of the product of the composite material's elastic modulus and the material thickness, the negative exponential of the ratio of ripple amplitude to material density, and the cosine function of the ripple tilt angle.

10. The biomimetic corrugated morphology design method for composite material functional structures according to claim 9, characterized in that, The underwater fluid experimental device includes an underwater fluid circulation system, a composite material specimen clamping device, a multi-dimensional force sensor, a flow velocity control system, and a data acquisition system. The underwater fluid circulation system provides a stable water flow environment with a flow velocity control range of 0.1–10.0 m / s and a flow velocity control accuracy of 0.01 m / s. The multi-dimensional force sensor measures the resistance and lift forces experienced by the specimen with a measurement accuracy of 0.001 N.

Images